Nvidia Requires Data Center Clients to Maintain Over 51% Non-China Business Share
nashnova research
Nvidia now requires global cloud clients to derive at least 51% of service revenue from non-Chinese customers — a move that narrows China's main workaround for accessing advanced AI chips through overseas data centers.
What exactly does the new rule demand?
Nvidia's sales teams have verbally told clients: non-China business must account for at least 51% of revenue.
Previously Nvidia placed no restrictions on who a client's customers were. The rule applies only to future contracts, not existing leases.
This means → Nvidia is extending control from "who can buy the chips" to "who the chips ultimately serve" — regulating not just hardware flow but the destination of compute power.
Why does this close China's workaround?
Many Chinese firms had been renting compute from overseas cloud providers — especially in Southeast Asia — to train large models, effectively sidestepping U.S. chip export controls.
Under the new rule, any cloud provider drawing more than half its revenue from Chinese clients risks losing access to Nvidia's chips.
In plain terms = the old restriction said "don't sell chips directly to China." The new one says "if your data center runs on our chips, it can't mainly serve Chinese clients either."
Why is Southeast Asia hit hardest?
A Singapore-based cloud provider told the Financial Times it would be extremely difficult to find enough non-Chinese clients outside the U.S., warning the rule "would definitely hurt business."
China and the U.S. are the world's two largest compute markets. Chinese tech firms were the biggest client group for Southeast Asian data centers.
This means → data centers that cannot quickly backfill with non-Chinese demand face pressure to scale down operations — idle capacity has no buyer.
How high have black-market chip prices climbed?
Nvidia's B300 — one of its most advanced chips, used by Anthropic and OpenAI — sells internationally at roughly $600,000 per rack.
Last month a rack traded in China for RMB 18.5 million (about $2.8 million) — nearly 4.7 times the international price.
This reflects a market where supply keeps tightening but demand has not faded — China's hunger for top-tier compute is spilling out as a black-market premium.
What is the cost to Nvidia itself?
Nvidia has already cut by more than half the number of authorized Asian chip buyers this year, imposing a new "whitelist" compliance regime.
Its stock has fallen more than 13% over the past month, losing the title of world's most valuable company to Apple — though it is still up about 7% year-to-date.
In plain terms = Nvidia is actively aligning with U.S. policy to tighten exports, but the cost shows up on the share-price chart — cutting clients means cutting revenue.
What is the key variable to watch next?
Nvidia's official response framed U.S.-chip-powered global data centers as "the foundation for American model and application developers to compete worldwide" — positioning itself as a gatekeeper of U.S. AI infrastructure, not just a chip supplier.
Whether Chinese tech firms can find alternative compute sources is the core variable that will test how effective this round of restrictions actually is.
This means → if substitutes mature, the controls lose bite; if none emerge, the training bottleneck for China's AI industry tightens further.
市场有风险，内容仅供研究参考，不构成投资建议。